Hi everyone,
I’m working on an AI-powered agent in Glide that reviews lift plans based on a structured review document. The agent needs to:
- Read the uploaded lift plan document.
- Check a predefined review document (which contains structured questions about the lift plan).
- Verify UK lifting regulations from a separate reference document.
- Answer each question in the review document based on the lift plan and regulations.
The problem is that my current setup requires 300+ updates per review, which makes it impractical due to Glide’s update limits.
I’m wondering if there’s a more efficient way to achieve this, maybe:
- A third-party AI service that integrates with Glide (e.g., OpenAI, Make, Zapier)?
- A way to batch-process the document analysis to reduce updates?
- Any alternative approaches that others have used for document review automation in Glide?
Has anyone tackled something similar, or does anyone have suggestions on optimizing this workflow?
Thanks in advance!
I suppose UK lifting regulation is what consumes the most updates here.
When you send a lot of information to AI together with a prompt (so called “retrieval augmented generation”) you would usually want to split the information into pieces digestible by AI and perform either a keyword or semantic search on the pieces (the latter will consume more tokens).
This way you only provide AI with pieces of information that are relevant to each question.
I recently built a RAG-style AI agent with Glide’s native query and workflows.
I assigned tags to pieces of documentation - in this case it was the related product.
I used “text to choice” to identify the tag related to the request.
Then I had three queries searching for related pieces of documentation.
Lastly, AI summarizes the content of the documentation and generates a response.
This way only the relevant documents are sent to to the model which saves a lot of tokens.
If you don’t have the patience to split the UK law into fragments yourself (which I totally understand), you might want to use some other RAG solutions outside of Glide.
There’s also the recent Pinecone integration which does something similar, although I haven’t had the chance to experiment with it yet.
2 Likes
Thanks for this insight! I also plan to try Pinecone when I have some time.
I think you should outsource this to Make to see if the updates count look better over there. Can you clarify which part costs the most updates? How are you configuring the “verify UK lifting regulations” step?